On Tue, May 26, 2020 at 02:08:41PM +1000, Stephen Rothwell wrote:
> Hi all,
>
> On Tue, 19 May 2020 15:09:55 +1000 Stephen Rothwell <s...@canb.auug.org.au>
> wrote:
> >
> > Hi all,
> >
> > After merging the drm-msm tree, today's linux-next build (arm
> > multi_v7_defconfig) failed like this:
> >
> > ERROR: modpost: "__aeabi_ldivmod" [drivers/gpu/drm/msm/msm.ko] undefined!
> > ERROR: modpost: "__aeabi_uldivmod" [drivers/gpu/drm/msm/msm.ko] undefined!
> >
> > Caused by commit
> >
> > 04d9044f6c57 ("drm/msm/dpu: add support for clk and bw scaling for
> > display")
> >
> > I applied the following patch for today (this is mechanical, there may
> > be a better way):
> >
> > From: Stephen Rothwell <s...@canb.auug.org.au>
> > Date: Tue, 19 May 2020 14:12:39 +1000
> > Subject: [PATCH] drm/msm/dpu: fix up u64/u32 division for 32 bit
> > architectures
> >
> > Signed-off-by: Stephen Rothwell <s...@canb.auug.org.au>
> > ---
> > drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c | 23 ++++++++++++++-----
> > drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c | 15 ++++++++----
> > 2 files changed, 28 insertions(+), 10 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
> > b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
> > index 9697abcbec3f..85c2a4190840 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
> > @@ -10,6 +10,7 @@
> > #include <linux/sort.h>
> > #include <linux/clk.h>
> > #include <linux/bitmap.h>
> > +#include <asm/div64.h>
> >
> > #include "dpu_kms.h"
> > #include "dpu_trace.h"
> > @@ -53,8 +54,11 @@ static u64 _dpu_core_perf_calc_bw(struct dpu_kms *kms,
> > }
> >
> > bw_factor = kms->catalog->perf.bw_inefficiency_factor;
> > - if (bw_factor)
> > - crtc_plane_bw = mult_frac(crtc_plane_bw, bw_factor, 100);
> > + if (bw_factor) {
> > + u64 quot = crtc_plane_bw;
> > + u32 rem = do_div(quot, 100);
> > + crtc_plane_bw = (quot * bw_factor) + ((rem * bw_factor) / 100);
> > + }
> >
> > return crtc_plane_bw;
> > }
> > @@ -89,8 +93,11 @@ static u64 _dpu_core_perf_calc_clk(struct dpu_kms *kms,
> > }
> >
> > clk_factor = kms->catalog->perf.clk_inefficiency_factor;
> > - if (clk_factor)
> > - crtc_clk = mult_frac(crtc_clk, clk_factor, 100);
> > + if (clk_factor) {
> > + u64 quot = crtc_clk;
> > + u32 rem = do_div(quot, 100);
> > + crtc_clk = (quot * clk_factor) + ((rem * clk_factor) / 100);
> > + }
> >
> > return crtc_clk;
> > }
> > @@ -234,8 +241,12 @@ static int _dpu_core_perf_crtc_update_bus(struct
> > dpu_kms *kms,
> > }
> > }
> >
> > - avg_bw = kms->num_paths ?
> > - perf.bw_ctl / kms->num_paths : 0;
> > + if (kms->num_paths) {
> > + avg_bw = perf.bw_ctl;
> > + do_div(avg_bw, kms->num_paths);
> > + } else {
> > + avg_bw = 0;
> > + }
> >
> > for (i = 0; i < kms->num_paths; i++)
> > icc_set_bw(kms->path[i],
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c
> > b/drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c
> > index c2a6e3dacd68..ad95f32eac13 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_plane.c
> > @@ -9,6 +9,7 @@
> >
> > #include <linux/debugfs.h>
> > #include <linux/dma-buf.h>
> > +#include <asm/div64.h>
> >
> > #include <drm/drm_atomic_uapi.h>
> > #include <drm/drm_damage_helper.h>
> > @@ -174,7 +175,11 @@ static void _dpu_plane_calc_bw(struct drm_plane *plane,
> > plane_prefill_bw =
> > src_width * hw_latency_lines * fps * fmt->bpp * scale_factor;
> >
> > - plane_prefill_bw = mult_frac(plane_prefill_bw, mode->vtotal, (vbp+vpw));
> > + {
> > + u64 quot = plane_prefill_bw;
> > + u32 rem = do_div(plane_prefill_bw, vbp + vpw);
> > + plane_prefill_bw = quot * mode->vtotal + rem * mode->vtotal /
> > (vbp + vpw);
> > + }
> >
> > pstate->plane_fetch_bw = max(plane_bw, plane_prefill_bw);
> > }
> > @@ -204,9 +209,11 @@ static void _dpu_plane_calc_clk(struct drm_plane
> > *plane)
> > pstate->plane_clk =
> > dst_width * mode->vtotal * fps;
> >
> > - if (src_height > dst_height)
> > - pstate->plane_clk = mult_frac(pstate->plane_clk,
> > - src_height, dst_height);
> > + if (src_height > dst_height) {
> > + u64 quot = pstate->plane_clk;
> > + u32 rem = do_div(quot, dst_height);
> > + pstate->plane_clk = quot * src_height + rem * src_height /
> > dst_height;
> > + }
> > }
> >
> > /**
> > --
> > 2.26.2
>
> I am still applying the above ...
>
> --
> Cheers,
> Stephen Rothwell
Additionally, I see a failure with clang due to the use of Bps_to_icc,
which does a straight division by 1000, which is treated as an integer
literal, with avg_bw as the dividend, which is a u64.
Below is the "hack" in my tree.
Cheers,
Nathan
diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
index 85c2a4190840..5ea725d8da6c 100644
--- a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
+++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_perf.c
@@ -250,7 +250,7 @@ static int _dpu_core_perf_crtc_update_bus(struct dpu_kms
*kms,
for (i = 0; i < kms->num_paths; i++)
icc_set_bw(kms->path[i],
- Bps_to_icc(avg_bw), (perf.max_per_pipe_ib));
+ div_u64(avg_bw, 1000), (perf.max_per_pipe_ib));
return ret;
}
